An oversampling method based on FEM for multibolt looseness monitoring using Lamb waves

Abstract

The damage states of multibolts are not only diverse but also difficult to obtain measured multibolt looseness signals due to time and economic costs. Therefore, the imbalance between healthy and damaged samples poses a challenge for monitoring multibolt looseness. A synthetic minority oversampling technique (SMOTE) based on finite element model is proposed to solve the problems of difficult acquisition of measured signals and imbalanced samples. A small amount of measurement data and a large amount of simulation data constitute the dataset, and their statistical characteristics are obtained. The spatial distribution of simulated samples representing the damage state is used as a constraint, and new samples are generated at the spatial positions of measured samples through an improved elliptical SMOTE method. To ensure high-quality sampling of new samples and achieve target distribution, an indicator based on Kullback–Leibler divergence was extracted. The new dataset combines the advantages of measurement data and simulation data to achieve a balance between health status data and damage status data, improving the quality of generated data. To verify the effectiveness of this method, a multibolt loosening monitoring platform is established, and the results showed that this method can improve the accuracy of early diagnosis of multibolt loosening.